Broadcom reinvents AMD's APU moniker as it launches Wi-Fi 8 chip — this accelerated processing unit doesn't include a GPU but has 10GbE connectivity

• Broadcom repurposes the APU label for networking silicon rather than graphics integration
• The BCM4918 shifts packet handling away from CPUs through dedicated offload engines
• Wi-Fi 8 access points increasingly resemble compact edge computing platforms

Broadcom has introduced the BCM4918 network processor for high-end Wi-Fi 8 residential access points, reviving the accelerated processing unit label in a context far removed from its original meaning.

Historically, the APU term described AMD processors that combined a general-purpose CPU with integrated graphics on one die.

In contrast, Broadcom applies the phrase to a system-on-chip that integrates compute cores, networking offload engines, security blocks, and on-device AI logic, without any GPU capability at all.

Compute and packet handling architecture

At the center of the BCM4918 sits a quad-core ARMv8-compatible CPU complex intended for control-plane operations and customer software.

Instead of handling traffic directly, the CPU is complemented by a dual-issue runner packet processor that manages wired and wireless data paths independently.

This design allows most network traffic to bypass the CPU entirely, reducing contention and avoiding software bottlenecks under sustained throughput demands.

Such separation between control and data planes is common in higher-end networking equipment, though its effectiveness in residential access points will depend on vendor firmware implementations.

Broadcom includes its Neural Engine as part of the BCM4918, enabling local inference for selected machine learning tasks.

This capability supports the idea of access points functioning as edge computing platforms rather than simple connectivity devices.

However, the available documentation does not quantify inference performance, supported models, or realistic workloads.

Without these details, the practical significance of on-device AI remains difficult to assess beyond general claims of autonomy and responsiveness.

The networking subsystem combines acceleration engines with integrated multi-gigabit Ethernet PHYs, including support associated with 10GbE Connectivity for wired backhaul scenarios.

Expansion options include four PCIe Gen3 interfaces and dual USB controllers, allowing additional radios or peripherals to be attached.

For security, features such as secure boot and cryptographic acceleration are built directly into the silicon, which should help residential networking hardware handle sensitive data and frequent software updates.

Broadcom emphasizes reduced board complexity by consolidating CPU cores, AI logic, networking acceleration, and security features into a single 19 x 19 mm FCBGA package for standard residential temperatures.

The BCM4918 appears less focused on current access point performance and more on future software-driven differentiation, assuming vendors can exploit capabilities that remain broadly described.

TechRadar will be extensively covering this year's CES, and will bring you all of the big announcements as they happen. Head over to our CES 2026 news page for the latest stories and our hands-on verdicts on everything from wireless TVs and foldable displays to new phones, laptops, smart home gadgets, and the latest in AI. You can also ask us a question about the show in our CES 2026 live Q&A and we’ll do our best to answer it.

And don’t forget to follow us on TikTok and WhatsApp for the latest from the CES show floor!